spark-plan-to-sql 0.1.0

Convert Apache Spark Catalyst LogicalPlan JSON back into a logically equivalent SQL statement.

spark-plan-to-sql

Convert Apache Spark Catalyst LogicalPlan JSON dumps back into a logically equivalent SQL statement.

The converter walks the pre-order JSON serialization Spark produces for its LogicalPlan (the same shape exposed via df.queryExecution.logical.toJSON) and emits readable SQL that, when re-executed, returns the same rows as the original query.

Install

pip install spark-plan-to-sql

Python API

import json
from spark_plan_to_sql import plan_to_sql, dict_to_sql

# 1) Plain JSON string
sql = plan_to_sql('[{"class": "...OneRowRelation", "num-children": 0}]')

# 2) Already-parsed Python list (Spark's native format)
with open("plan.json") as f:
    sql = plan_to_sql(json.load(f))

# 3) A dict wrapper, e.g. {"plan": [...]} or {"logicalPlan": [...]}
sql = plan_to_sql({"plan": json.load(open("plan.json"))})

# 4) Strict dict-only helper
sql = dict_to_sql({"logicalPlan": [...]})

The function accepts:

• str / bytes — JSON text
• list[dict] — Spark's native pre-order plan list
• dict — either a single leaf node or a wrapper such as {"plan": [...]} / {"logicalPlan": [...]} / {"nodes": [...]}

CLI

# Convert one or more files (prints to stdout)
spark-plan-to-sql plan.json plan2.json

# Batch convert a directory of plans into ./restored_sql/
spark-plan-to-sql --dir test_json --out restored_sql

Supported plan nodes

DDL/DML: CreateNamespace, DropNamespace, SetCatalogAndNamespace, CreateTable*, DropTable*, DropView, CreateViewCommand, CacheTable, UncacheTable, InsertInto*, AppendData.

Query: Project, Filter, Sort, GlobalLimit/LocalLimit, Distinct/Deduplicate, Aggregate (incl. ROLLUP/CUBE/GROUPING SETS through the Expand+spark_grouping_id pattern), Join (Inner / Left / Right / Full / LeftSemi / LeftAnti / Cross), Union/Intersect/Except, Window (with frame suppression for LAG/LEAD/RANK/ROW_NUMBER/...), Generate (LATERAL VIEW), WithCTE/CTERelationDef/CTERelationRef, SubqueryAlias, LogicalRelation, DataSourceV2Relation, LocalRelation, OneRowRelation.

Expressions: literals (incl. interval/decimal/date/timestamp), Cast, Alias, AttributeReference, OuterReference, binary/unary operators, CaseWhen/If, Coalesce/IfNull/Nvl, aggregates (Count/Sum/ Avg/...), datetime (Year/Month/AddMonths/DateAdd rewriting from ExtractANSIIntervalDays/...), windowing (WindowExpression/ WindowSpecDefinition/SpecifiedWindowFrame), array/map/struct (GetStructField/GetArrayItem/ElementAt/ArrayTransform/ LambdaFunction/CreateNamedStruct/MapFromArrays/...), JSON (JsonToStructs/GetJsonObject/StructsToJson).

License

MIT